Strategies for handling dose effects in network meta-analysis: a review of practice and methodology




Poster session 3


Tuesday 25 October 2016 - 10:30 to 11:00


All authors in correct order:

Yuan J1, Caldwell D2, Mao C1, Tang J1, Higgins J2
1 Cochrane Hong Kong, CUHK, Hong Kong
2 School of Social and Community Medicine, University of Bristol, UK
Presenting author and contact person

Presenting author:

Chen Mao

Contact person:

Abstract text
Background: Dose effect is a common and important issue in network meta-analyses of pharmaceutical interventions but the methodology for this issue has received relatively little attention.

Objectives: To summarize strategies used in practice and propose a methodological framework specifically for handling dose effects in network meta-analyses.

Methods: We systematically reviewed published network meta-analyses with four or more intervention nodes, of which at least one was a pharmaceutical intervention. Strategies used for addressing dose effects were summarized. Methodology papers (dose effects in pairwise meta-analysis, model-based meta-analysis, and modeling dose in network meta-analysis) were also reviewed though this was not performed systematically.

Results: The review of practice was based on 350 network meta-analyses. We identified 76 (21.7%) network meta-analyses which did not report any drug dose information, and 93 (26.6%) network meta-analyses involving drugs with multiple doses but in which the potential effects were not appropriately addressed. We found 166 (47.4%) network meta-analyses applying one or more specific strategies, including restricting attention to specific doses (58 studies), splitting doses (87), lumping multiple doses with supporting evidence (24), stratified analysis by dose (5), modeling dose-response (2), and unspecified meta-regression (2). We propose a methodological framework for addressing dose effects, which combines methodological considerations with strategies used in practice.

Conclusion: Dose effects were often not handled appropriately in published network meta-analyses, although a number of useful strategies are available. Our proposed framework specifically for handling dose effects will hopefully be useful for future network meta-analysis authors.